A survey on unsupervised learning for wearable sensor-based activity recognition

Human Activity Recognition (HAR) is an essential task in various applications such as pervasive healthcare, smart environment, and security and surveillance. The need to develop accurate HAR systems has motivated researchers to propose various recognition models, feature extraction methods, and datasets. A lot of comprehensive surveys have been done on vision-based HAR, while few surveys have been done on sensor-based HAR. The few existing surveys on sensor-based HAR have focused on reviewing various feature extraction methods, the adoption of deep learning in activity recognition, and existing wearable acceleration sensors, among other areas. In recent times, state-of-the-art HAR models have been developed using wearable sensors due to the numerous advantages it offers over other modalities. However, one limitation of wearable sensors is the difficulty of annotating datasets during or after collection, as it tends to be laborious, time-consuming, and expensive. For this reason, recent state-of-the-art activity recognition models are being proposed using fully unlabelled datasets, an approach which is described as unsupervised learning. However, no existing sensor-based HAR surveys have focused on reviewing this recent adoption. To this end, this survey contributes by reviewing the evolution of activity recognition models, collating various types of activities, compiling over thirty activity recognition datasets, and reviewing the existing state-of-the-art models to leveraging fully unlabelled datasets in activity recognition. Also, this survey is the first attempt at a comprehensive review on the adoption of unsupervised learning in wearable sensor-based activity recognition. This will give researchers in this area a solid background and knowledge of the existing state-of-the-art models and an insight into the grand research areas that can still be explored.